The prediction of energy conversion during the self-propelled jumping of multidroplets based on convolutional neural networks

The energy conversion efficiency (the ratio of the maximum jumping kinetic energy to the maximum surface energy released from droplet coalescence) is an essential indicator of the self-propelled jumping of droplets, which determines its value for applications in various fields. In the practical condensation process, the initial states of the multidroplets with different sizes and distributions have a significant effect on the energy conversion efficiency, but the mechanism behind this effect remains unclear. This paper reveals the effect of the initial states of droplets on the energy conversion efficiency of multidroplet jumping (mainly three droplets) from the perspective of energy conversion and the internal flow of the merged droplets. Different initial states will lead to different flow directions of the liquid microclusters inside the merged droplets. The consistency between the flow direction and the jumping direction will affect the energy conversion efficiency. To characterize this effect quantitatively, we construct a machine learning model based on a convolutional neural network to predict the energy conversion efficiency of multidroplet jumping with different initial distribution angles and radius ratios. The input of the neural network is the images of the initial state of the droplets, and the output is the energy conversion efficiency. After training, the neural network can predict the energy conversion efficiency of multidroplet jumping with an arbitrary initial state.